How can I use spectroscopy to predict the amount of a substance?

[#1]

Hello, I have a rather generic spectroscopy question, I’m hoping someone can advise me on how to use JMP.

The analytical chemist wants to predict the ppb of a substance, based on the spectroscopy profile. 

I fit a Multivariate Curve Resolution model and it fit well. This option is available in Functional Data Explorer.

How can this model be used in the future to predict the ppb, given a table of excitation/wavelength values from spectroscopy?

Thanks, Paul

Model

Raw Data

[#2 · Accepted solution]

Hi @paulp,

You might find my previous answer to the same topic in this post :Functional Data Analysis and Classification: How to calculate FPC of new data (unseen by the origina...

Basically the new rows/data can be entered as validation or test rows, so that your original model remains the same, and so you could calculate the predicted response for the validation/test rows.

Note that if you're using Functional DOE Analysis, you have an option to save Prediction formula directly in your table : 

Hope this answer will help you,

[#3]

Hello Victor, thank you for your reply.  Saving the Prediction Formula did work with the Fourier Basis, and the fit was good.   

Initially the model used in FDE was Multivariate Curve Resolution which also fit well - but seems it doesn't provide the Prediction Formula. When I save the Prediction Formula, that new column appears to be empty in the data table. 

[#4]

Hi @paulp,

Not directly related to your question, but did you try the Wavelets models family for your spectroscopic data ?

If not, I would considere giving it a try, as it performs quite well on spectral data.

Concerning the MCR, I can't help you since I don't seem to have this option (yet), I'm with JMP Pro 17.2.0 version and MCR seems to be an option from JMP 18.

Maybe someone from JMP technical staff could help you figure out where does the error comes from : @Bill_Worley, @JeremyAshJMP @chris_gotwalt1 ?